[vLLM IR] 1/N Implement IR skeleton and rms_norm op (#33825)

Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Signed-off-by: Xinyu Chen <xinyu1.chen@intel.com>
Signed-off-by: chzhang <chaojun.zhang@intel.com>
Signed-off-by: Luka Govedic <luka.govedic@gmail.com>
Co-authored-by: Xinyu Chen <xinyu1.chen@intel.com>
Co-authored-by: Chaojun Zhang <chaojun.zhang@intel.com>
Co-authored-by: Luka Govedič <ProExpertProg@h100-01.nemg-001.lab.rdu2.dc.redhat.com>
This commit is contained in:
Luka Govedič
2026-03-31 22:15:05 -04:00
committed by GitHub
co-authored by Xinyu Chen Chaojun Zhang Luka Govedič
parent 0fab52f0aa
commit 40bb175027
49 changed files with 2177 additions and 265 deletions
+11
View File
@@ -17,6 +17,7 @@ if TYPE_CHECKING:
from torch.distributed import PrefixStore, ProcessGroup
from vllm.config import VllmConfig
from vllm.config.kernel import IrOpPriorityConfig
from vllm.inputs import EngineInput
from vllm.pooling_params import PoolingParams
from vllm.sampling_params import SamplingParams
@@ -931,6 +932,16 @@ class Platform:
"num_compute_units is not implemented for the current platform."
)
@classmethod
def get_default_ir_op_priority(
cls, vllm_config: "VllmConfig"
) -> "IrOpPriorityConfig":
"""Get the default IR op priority for the current platform."""
from vllm.config.kernel import IrOpPriorityConfig
# Native always used by default. Platforms can override this behavior.
return IrOpPriorityConfig.with_default(["native"])
class UnspecifiedPlatform(Platform):
_enum = PlatformEnum.UNSPECIFIED